EDIT Sound Mar22_000_PSI_mar15 28/02/2022 16:51 Page 1
SECURITY
Sound advice for CCTV The use of automatic sound recognition can improve a surveillance installation and help reduce unwanted alerts. i-PRO has released a white paper discussing the subject
T
he surveillance camera industry is promoting greater use of AI-based video analysis technology. With a surveillance camera and an AI video analysis app, suspicious individuals can be automatically detected and the individual's physical characteristics can be used to perform a video library search. This reduces the monitoring burden placed on the camera operator and improves efficiency. Further surveillance efficiency can be realised by analysing the camera's audio data together with the video data. Especially in a noisy environment, it is difficult for an operator to listen to sound captured by multiple cameras at the same time. Therefore, automatic audio analysis technology also offers a lot of promise. One of the sound detection features of conventional surveillance cameras is a function that issues an alarm whenever the volume exceeds a specific level. In a noisy environment, this leads to many false alarms, putting considerable restrictions on audio surveillance at a practical level. Just like with AI-based video analysis, by using AI sound classification technology to detect abnormal sounds such as people yelling, we can detect incidents earlier and search recorded video more effectively.
Edge AI processing Read the white paper here:
40
In the past, AI-based video analysis usually took the form of a server-based system that performed arithmetic processing on video uploaded to the server. In recent years however, Edge AI, which allows for Deep Learning arithmetic processing to be performed within each surveillance camera, is becoming widespread. The advantages of Edge AI include cost reduction, better information security
and enhanced real-time performance. By installing various apps on an IP camera, customers can easily adopt video analysis technology with Edge AI. With an i-PRO network camera, for example, both video and audio analyses are performed by Edge AI processing. Audio analysis generally involves a smaller amount of data than for video analysis and can be processed relatively easily. This means both video and audio analyses can be performed simultaneously within the camera. The AI sound classification technology uses two indicators, the captured sound volume level and an AI score, to determine whether an alarm should be issued.
Audio analysis generally involves a smaller amount of data than for video analysis and can be processed relatively easily. This means both video and audio analyses can be performed simultaneously within the camera To identify a sound, the system first compares the captured sound volume level with a preset threshold value. If it is greater than the threshold, AI is then used to determine what kind of sound it could be. To come up with an AI score, the system determines whether the captured sound corresponds to any of four target sound categories: yell, glass break, vehicle horn, and gunshot. To do this, the captured sound is divided into regular segments, signal processing is performed, and the feature quantity for the target sound is determined. By inputting the feature quantity into
www.psimagazine.co.uk
